Question
Which AI-system roles are better served by small language models than by frontier general models?
Hypothesis
Small models can be strong at bounded classification, routing, extraction, and critique when the task contract is tight.
Method
Define narrow tasks, measure latency and accuracy, then compare against larger-model baselines with equal prompts and datasets.
Prototype
Create a router that selects retrieval strategy, tool policy, or model tier based on user intent and confidence thresholds.
Notes
The goal is not novelty for its own sake. The goal is product-quality behavior with less waste.
Results / Open Questions
Open question: how much calibration data is needed before a small model earns production trust?
References
Placeholder for SLM, distillation, model routing, and edge inference reading.